Research map / 3D vision / 3D generation
MVTN: Learning Multi-view Transformations for 3D Understanding
Abdullah Hamdi, Faisal AlZahrani, Silvio Giancola, Bernard Ghanem
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add MVTN view regression | ViewGCN (12 views, same renderer) · ModelNet40 | overall accuracy: 93.0 -> 93.8 | Improved |
| add MVTN view regression | ViewGCN (20 views, same renderer) · ModelNet40 | per-class accuracy: 91.3 -> 92.2 | Improved |
| add learned MVTN-spherical views instead of fixed spherical (ablation) | MVCNN · ModelNet40 | test accuracy: 92.8 -> 93.3 | Improved |
| use random object color in training (ablation) | MVTN · ModelNet40 | test accuracy: 93.3 -> 93.4 | No clear effect |
| use deeper ResNet-34 backbone (ablation) | MVTN-spherical (PointNet) · ModelNet40 | test accuracy: 93.41 -> 92.83 | No clear effect |
| use DGCNN instead of PointNet point encoder (ablation) | MVTN-spherical (ResNet-18) · ModelNet40 | test accuracy: 93.41 -> 93.26 | No clear effect |
| use late fusion instead of MVTN pipeline (ablation) | MVTN · ModelNet40 | accuracy: 91.8 -> 88.4 | Hurt |
| use MVTN-circular views under rotation perturbation | MVTN with MVCNN · ModelNet40 rotation +-180 deg | overall accuracy: 90.3 -> 91.2 | Improved |
| add MVTN under 50 percent occlusion | MVTN vs PointNet · ModelNet40 occlusion 0.5 | accuracy: 53.5 -> 67.1 | Improved |
| add ImageNet pretraining (ablation) | MVTN with ViT · ScanObjectNN hardest | overall accuracy: 61.97 -> 84.21 | Improved |
| add ImageNet pretraining (ablation) | MVTN with ResNet-50 · ScanObjectNN hardest | overall accuracy: 61.21 -> 81.26 | Improved |
| use full-supervised pretraining on single views (ablation) | MVTN with ViT · ScanObjectNN hardest | overall accuracy: 61.97 -> 57.53 | Hurt |
Closest papers by meaning
- MVTN: Multi-View Transformation Network for 3D Shape Recognition (ICCV 2021 · similarity 0.98)
- MVImgNet: A Large-Scale Dataset of Multi-View Images (CVPR 2023 · similarity 0.90)
- Learning View-Based Graph Convolutional Network for Multi-View 3D Shape Analysis (TPAMI 2022 · similarity 0.89)
- MLVCNN: Multi-Loop-View Convolutional Neural Network for 3D Shape Retrieval (AAAI 2019 · similarity 0.89)
- View-GCN: View-Based Graph Convolutional Network for 3D Shape Analysis (CVPR 2020 · similarity 0.88)
- Duoduo CLIP: Efficient 3D Understanding with Multi-View Images (ICLR 2025 · similarity 0.87)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
